Andrew Ng's AI Engineering Skills Map: The 4 Skills That Matter in 2026

On August 14, 2026, Andrew Ng published the AI Engineering Skills Map — a synthesis of over 10,000 job postings, dozens of structured interviews with AI experts and hiring managers, and survey data. The goal: help developers prioritize what to learn and help employers hire skilled developers.

The result is a clear, four-skill framework that cuts through the noise of the AI hype cycle. And it comes with a critical insight: “AI Engineering skills” is not the same as the “AI Engineer” role. Just as all developers need cloud skills (not just those with “Cloud Engineer” titles), all developers will need AI engineering skills.

The 4 Essential AI Engineering Skills

Skill 1: Building and Deploying AI Applications

The key difference between AI and non-AI applications: AI outputs are unpredictable. When you prompt an LLM, you don’t know what you’ll get back. When you train a deep learning model, you don’t know what prediction it will make on new examples. Traditional software behaves more predictably.

People skilled at building and deploying AI applications understand the building blocks — LLMs, context engineering, RAG, agentic workflows, machine learning and deep learning — and, crucially, how to use statistical techniques to measure, steer, and govern AI systems so they behave more predictably.

The core skill within this: driving disciplined evals and error analysis loops. Building an AI demo is easy. Building an AI system that behaves reliably in production is hard. The difference is evaluation discipline.

Skill 2: Software Engineering Fundamentals

When you deeply understand how software works, you build more effectively. Engineering software requires making tradeoffs between cost, scalability, reliability, speed, security, and privacy.

Understanding software fundamentals lets you recognize what tradeoffs exist — leading to better decisions in choosing your software stack, designing system architecture, designing data stores, and testing. It also leads to much better outcomes than an inexperienced developer who “vibe codes” a solution without knowing the tradeoffs their coding agent is making.

As Ng notes: “Understanding software engineering fundamentals lets you make good tradeoffs by steering coding agents using the precise language of software engineering.” Without fundamentals, you can’t effectively guide the AI tools that are now writing much of the code.

Skill 3: Using Coding Agents

Using agentic coding effectively is now a key skill for every developer. When you have this skill, you have a good mental model for how agents work, understand their limitations, know how to work around them, and can quickly steer them — knowing how much to intervene and how much to leave alone.

This requires managing a coding agent’s context, making tradeoffs about what to include and exclude, and steering the agent with the precise language of software engineering. The agent is a tool. You’re the engineer.

The InfoQ analysis of Microsoft’s Agent Framework GA reinforces this: the harness (the runtime that manages the agent loop, tool invocation, context, and recovery) is where the engineering happens. An April 2026 paper from MBZUAI’s VILA-Lab analyzed Claude Code and found that 98.4% of the codebase is harness infrastructure — permissions, context management, sandboxing, tool routing — and only 1.6% is AI decision logic. The skill isn’t in the AI. It’s in the engineering around the AI.

Skill 4: Shaping the Build

The fourth skill is guiding AI development with domain expertise and evaluation. This means knowing what “good” looks like in your specific domain, designing evaluations that catch failures, and iterating on quality — not just functionality.

This is where senior engineers separate from juniors. A junior can get an AI feature working. A senior can tell you whether it’s good enough for production, what its failure modes are, and how to improve it. Shaping the build is the skill that makes the other three valuable.

The Market Reality

The job market data confirms what the skills map describes:

AI skills are now mainstream requirements. ZipRecruiter’s 2026 AI Employer Report found that 74% of employers see AI skills as a strong advantage or outright requirement for at least some roles, with 13% stating AI skills are required for all roles. Dice reports AI skill requirements in 79% of US tech job postings in July 2026 — up 144% year-over-year.

The senior-junior gap is widening. The AI Pulse Q2 2026 report found:

• Senior IC roles (5+ years): 52% year-over-year growth
• Mid-level IC roles (3-5 years): 28% growth
• Junior IC roles (0-2 years): only 11% growth

AI is automating the routine work that junior engineers used to do. Companies are hiring fewer juniors and asking the seniors they have to do more. The skills that differentiate seniors — software fundamentals, evaluation discipline, shaping the build — are exactly the skills Ng’s map identifies.

Compensation reflects the demand. Senior AI engineers with 5+ years and shipped AI features earn $300K-$450K base, $500K-$800K total comp including equity at AI-native scale-ups. Research scientists at top labs earn $700K-$1.5M+ total comp. AI product managers earn $250K-$400K base with significant equity.

Pure prompt engineer roles are declining. The AI Pulse report notes that pure Prompt Engineer titles are becoming less common — the role is being absorbed into AI Systems Engineering, with prompt engineering as one skill within a broader role. The Dice report confirms this: year-over-year skills growth is in Agentic AI, AI Agents, Responsible AI, and AI Infrastructure — not prompt engineering in isolation.

The Skills That Employers Value Most

ZipRecruiter found that employers value both AI-specific and human skills more than a year ago:

AI-specific skills (growing in importance):

• Workflow automation: 60% of employers say more important
• Data analysis: 60%
• AI governance: 56%

Human skills (growing in importance):

• Critical thinking: 65%
• Judgment and decision-making: 59%
• Creativity: 58%

The Dice report’s fastest-growing skills tell the same story: Organizational Change Management (42% MoM growth), Quality Improvement (37%), Operational Performance Management (24%), Responsible AI (21%). These aren’t coding skills — they’re the skills needed to deploy AI in real organizations.

The message: the most valuable AI engineer isn’t the one who can build the best model. It’s the one who can deploy AI responsibly, evaluate it rigorously, and navigate the organizational change it requires.

What This Means for Professionals

1. Don’t Chase the Title. Build the Skills.

The “AI Engineer” job title is still emerging and inconsistently defined. What’s not emerging is the demand for AI engineering skills. Whether you’re a full-stack engineer, data engineer, DevOps engineer, or ML engineer, you need these four skills. Build them within your current role — don’t wait for a title change.

2. Evals Are the Differentiator

Anyone can call an API. The skill that separates competent AI engineers from dangerous ones is evaluation discipline. Learn to design evals, run error analysis, and determine whether an AI system is production-ready. This is Skill 1 (building/deploying) and Skill 4 (shaping the build) combined.

3. Software Fundamentals Matter More, Not Less

The rise of coding agents doesn’t eliminate the need for software engineering fundamentals — it amplifies it. Without fundamentals, you can’t steer coding agents effectively. You can’t evaluate their output. You can’t make architecture tradeoffs. The agents make decisions for you, and those decisions are often poor.

4. Learn to Steer Coding Agents

This is the newest skill on the list and the one most developers haven’t systematically developed. Practice managing agent context, knowing when to intervene, and steering with precise instructions. The developers who master this are 5-10x more productive than those who don’t.

The Executive AI Workshop includes hands-on sessions on coding agent productivity: teaching teams how to steer agents effectively, design evals, and build the harness infrastructure that makes AI reliable in production.

What This Means for Hiring Managers

1. Hire for the Four Skills, Not the Title

Don’t filter for “AI Engineer” in job titles. Filter for evidence of the four skills: deployed AI applications, software engineering depth, coding agent proficiency, and evaluation discipline. A senior full-stack engineer who’s built and deployed RAG systems is more valuable than someone with an “AI Engineer” title who’s only fine-tuned models in notebooks.

2. The Junior Pipeline Needs Redesign

Junior AI roles are growing at only 11% YoY. If you’re not hiring juniors, you’re not building the pipeline for future seniors. Redesign entry-level roles to develop the four skills — give juniors eval design work, coding agent steering practice, and deployment experience, not just model training tasks.

3. Training Is Your Responsibility

Only 22% of employers provide mandatory AI training for all employees (ZipRecruiter). 17% provide none at all. If you’re not training your workforce in the four skills, you’re relying on them to learn independently — which means your competitors who do train will have a structural advantage.

The AI Agent Automation Consulting service includes team capability assessments: evaluating your current team against the four-skill framework and designing a training plan to close the gaps.

The India Context

The Dice report’s fastest-growing skills — Organizational Change Management (42%), Responsible AI (21%), Business Transformation (19%) — align closely with Ng’s framework. But the Deloitte 2026 State of AI report found that India reports lower levels of AI expertise (0-4%) compared to other countries (2-8%).

This is both a challenge and an opportunity. The challenge: India’s AI adoption is outpacing its AI skill development. The opportunity: the four-skill framework provides a clear roadmap for closing the gap. Indian developers who build these four skills — particularly evaluation discipline and coding agent proficiency — will be disproportionately valuable in a market where these skills are scarce.

The AI Strategy for Business consultation for Indian organizations now includes a Skills Gap Assessment: mapping your team’s current capabilities against Ng’s four-skill framework and building a prioritized development plan.

The Bottom Line

Andrew Ng’s AI Engineering Skills Map is the clearest framework we have for what it takes to be an AI engineer in 2026. Four skills. Not a job title. Not a certification. Not a degree. Four skills that can be developed by any developer willing to put in the work.

The market data confirms the framework: 74% of employers require AI skills. Senior AI engineers earn $500K-$800K. Pure prompt engineer roles are declining. The gap between juniors and seniors is widening. The skills that matter — evaluation discipline, software fundamentals, coding agent proficiency, shaping the build — are exactly what Ng identified.

The message for professionals: don’t chase the “AI Engineer” title. Build the four skills. The titles will follow the skills, not the other way around.

The message for employers: hire for the four skills, train for the four skills, and redesign your junior pipeline to develop them. The organizations that build these capabilities across their engineering teams will deploy AI faster, more reliably, and more cost-effectively than those that rely on a few specialists with the right job title.

AI engineering isn’t a role. It’s a skill set. And it’s the skill set that every developer needs to build.